Sentence generating method and apparatus

1. A sentence generating method, comprising:

generating, by at least one processor, a corresponding word set of a source word set generated based on a source sentence, wherein the source sentence and the source word set are in a first language, and wherein the corresponding word set is in a second language different from the first language;

generating, by the at least one processor, words in the second language by performing decoding based on feature vectors generated through encoding of the source sentence;

adjusting, by the at least one processor, a probability of at least one of the generated words based on either one or both of the source word set and the corresponding word set;

selecting, by the at least one processor, based on the adjusted probability and the probability as unadjusted, character strings from different character strings including each of the generated words; and

generating, by the at least one processor, a translated sentence of the source sentence based on the selected character strings.

2. The sentence generating method of claim 1 , wherein the adjusting of the probability comprises, in response to a word identified among the generated words being included in the corresponding word set, adjusting a probability of the identified word.

3. The sentence generating method of claim 1 , wherein the adjusting of the probability comprises,

in response to a word identified among the generated words not being included in the corresponding word set, verifying whether a common word is present among other words in a group including the identified word and words in the corresponding word set, and

in response to the common word being verified as present and the common word being one of the generated words, adjusting a probability of the common word.

4. The sentence generating method of claim 1 , wherein the adjusting of the probability comprises

identifying a word among the generated words based on a probability of each of the generated words, and

adjusting the probability of the identified word, in response to the identified word being included in the corresponding word set.

5. The sentence generating method of claim 1 , wherein the performing of the decoding comprises

calculating a context vector using the feature vectors and weights, and

generating the words based on the calculated context vector and a previous decoding result.

6. The sentence generating method of claim 1 , wherein the adjusting of the probability comprises

identifying a source word corresponding to a word having a maximum probability among the generated words based on a maximum value among weights used to perform the decoding, wherein the identified source word is one of words in the source sentence,

retrieving a corresponding word of the identified source word from the corresponding word set, in response to the identified source word being included in the source word set; and

adjusting a probability of the retrieved corresponding word, in response to the retrieved corresponding word being one of the generated words.

7. The sentence generating method of claim 6 , wherein the adjusting of the probability comprises,

in response to the probability of the retrieved corresponding word being adjusted, identifying a word among the retrieved corresponding word and a remaining word among the generated words excluding the retrieved corresponding word, based on a probability of the remaining word and the adjusted probability of the retrieved corresponding word, and

in response to the identified word being included in the corresponding word set, adjusting a probability of the identified word.

8. The sentence generating method of claim 1 , further comprising, in response to the selected character strings being candidates for the translated sentence:

calculating a score of each of the selected character strings;

applying a coverage of each of the selected character strings to the score of a respective one of the selected character strings; and

determining one of the selected character strings to be the translated sentence based on a result of the applying of the coverage of each of the selected character strings,

wherein the coverage of each of the selected character strings indicates how many words in the corresponding word set are included in each of the selected character strings.

9. The sentence generating method of claim 1 , wherein

the source word set comprises any one or any combination of any two or more of a word extracted from the source sentence, a synonym of the extracted word, and a basic form of the extracted word, and

the corresponding word set comprises any one or any combination of any two or more of a translated word obtained by representing a word in the source word set in the second language, a synonym of the translated word, and a basic form of the translated word.

10. The sentence generating method of claim 1 , wherein the corresponding word set is generated using a first language-second language dictionary as a reference.

11. A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the sentence generating method of claim 1 .

12. A sentence generating apparatus, comprising:

a controller configured to

generate a corresponding word set of a source word set generated based on a source sentence, wherein the source sentence and the source word set are in a first language, and wherein the corresponding word set is in a second language different from the first language,

generate words in the second language by performing decoding based on feature vectors generated through encoding of the source sentence,

adjust a probability of at least one of the generated words based on either one or both of the source word set and the corresponding word set,

select, based on the adjusted probability and the probability as unadjusted, character strings from different character strings including each of the generated words, and

generate a translated sentence of the source sentence based on the selected character strings.

13. The sentence generating apparatus of claim 12 , wherein the controller is further configured to adjust a probability of an identified word among the generated words, in response to the identified word being included in the corresponding word set.

14. The sentence generating apparatus of claim 12 , wherein

the controller is further configured to verify whether a common word is present among other words in a group including an identified word, among the generated words, and words in the corresponding word set, in response to the identified word not being included in the corresponding word set, and

the controller is further configured to adjust a probability of the common word, in response to the common word being verified as present and the common word being one of the generated words.

15. The sentence generating apparatus of claim 12 , wherein the controller is further configured to

identify a word among the generated words based on a probability of each of the generated words, and

adjust the probability of the identified word, in response to the identified word being included in the corresponding word set.

16. The sentence generating apparatus of claim 12 , wherein the controller is further configured to

calculate a context vector using the feature vectors and weights, and

generate the words based on the calculated context vector and a previous decoding result.

17. The sentence generating apparatus of claim 12 , wherein the controller is further configured to

identify a source word corresponding to a word having a maximum probability among the generated words, based on a maximum value of weights used to perform the decoding, wherein the identified source word is one of words in the source sentence,

retrieve a corresponding word of the identified source word from the corresponding word set, in response to the identified source word being included in the source word set, and

adjust a probability of the retrieved corresponding word, in response to the retrieved corresponding word being one of the generated words.

18. The sentence generating apparatus of claim 17 , wherein

the controller is further configured to identify a word among the retrieved corresponding word and a remaining word among the generated words excluding the retrieved corresponding word, in response to the probability of the retrieved corresponding word being adjusted, based on a probability of the remaining word and the adjusted probability of the retrieved corresponding word, and

the controller is further configured to adjust a probability of the identified word, in response to the identified word being included in the corresponding word set.

19. The sentence generating apparatus of claim 12 , wherein the controller is further configured to, in response to the selected character strings being candidates for the translated sentence,

calculate a score of each of the selected character strings,

apply a coverage of each of the selected character strings to the score of a respective one of the selected character strings, and

determine one of the selected character strings to be the translated sentence based on a result of the applying of the coverage of each of the selected character strings,

wherein the coverage of each of the selected character strings indicates how many words in the corresponding word set are included in each of the selected character strings.

20. The sentence generating apparatus of claim 12 , wherein

the source word set comprises any one or any combination of any two or more of a word extracted from the source sentence, a synonym of the extracted word, and a basic form of the extracted word, and

the corresponding word set comprises any one or any combination of any two or more of a translated word obtained by representing a word in the source word set in the second language, a synonym of the translated word, and a basic form of the translated word.

21. The sentence generating apparatus of claim 12 , wherein the corresponding word set is generated using a first language-second language dictionary as a reference.